Inside ChatGPT's rushed launch, as some OpenAI executives feared rivals upstaging them; Sam Altman worries about a regulatory backlash and inflated expectations
The New York Times' account lands at the start of a longer arc its own coverage later confirmed: the rushed November 2022 launch that some executives greenlit out of fear of being upstaged is what triggered a year-long scramble across Silicon Valley, pulling Google, Meta, and the rest into a catch-up race they had not planned for.
OpenAI's competitive fear set product policy: a research demo shipped early as a consumer product, immediately exposing the company to the public scrutiny and regulator attention Altman says he feared.
Google and Meta were forced onto the back foot, responding to a launch they did not initiate rather than setting the agenda for conversational AI themselves.
Second-order effects
The scramble legitimized ChatGPT as a platform, which pushed Altman toward the personal-assistant-for-work pitch — a move that put OpenAI in direct competition with its closest partner, Microsoft.
Rapid post-launch hiring (80% of the current staff joined after the release) meant the company scaling the product barely overlapped with the team that built and debated it, straining the culture as it pivoted commercial.
Third-order effects
If competitive fear keeps overriding deliberation at the frontier labs, release decisions get made by rivalry rather than risk assessment — the pattern regulators cite when arguing frontier-model deployment needs external oversight.
Altman's expectation-management problem compounds structurally: each launch raises the bar the next one must clear, tying OpenAI's credibility to sustaining growth signals like the 10%-monthly-growth messaging he later sent staff.
The trend: ChatGPT's rushed launch marks the moment frontier labs began letting competitive timing, not readiness, dictate when powerful models reach the public.
Data on the intellectual contribution to AI from various research organizations. Some of organizations publish knowledge and open-source code for the entire world to use. Others just consume it. https://twitter.com/...
To be fair, @openAI had a big impact on AI/ML. Their OpenAI Gym env for RL, their CLIP - a de facto model for any image-text research, their Whisper speech recognition is amazing - all open sourced. In my view, they've had a bigger impact than many labs publishing 100s papers 🧵 h…
OpenAI has been closely guarding details about ChatGPT's popularity, saying only that it has more than a million users. The real numbers: two months after launch, it has more than 30 million users, and gets roughly 5 million visits a day. https://www.nytimes.com/...
“One day in mid-November, workers at OpenAI got an unexpected assignment: Release a chatbot, fast. The chatbot, an executive announced, would be known as “Chat with GPT-3.5,” and it would be made available free to the public. In two weeks.” kudos @sama. https://www.nytimes.com/..…
And when I think of “intellectual contributions” to AI research, I'd guess that much of the most important work isn't on arXiv at all. It's in books or journals that many computer scientists seem unwilling to learn about (or take the time to read). >>
@jhoang314 Hate? Where? We do work together. OpenAI uses PyTorch, which was developed at FAIR. PyTorch 2.0 uses the Triton back-end compiler which was developed at OpenAI. OpenAI use transformers and RLHF which originated at Google & DeepMind.
ChatGPT is, for now, a money pit, @kevinroose notes. There are no ads, and the average conversation costs OpenAI “single-digit cents” in processing power, according to a post on Twitter by Sam Altman, the company's CEO. https://www.nytimes.com/...
We know Meta's head of AI is right about GPT not being a big deal because: 1. GPT isn't the fastest growing consumer app in history, and 2. Google didn't call a code red because they're worried about the survival of their business
@jhoang314 But the fact is that most of the ideas, techniques and tools used by OpenAI came from Google, FAIR & DeepMind. That's totally fine. But OpenAI products do not come out of a vacuum. Like every new product, they integrate existing technology from other orgs and add a coa…
@DanielMiessler I wouldn't necessarily count a business ignoring a potential threat as proof it isn't a big deal. That would just another headstone in a big graveyard of ignorant companies in history.
And yes it's a problem that OpenAI is not being transparent about what they are unleashing on the world. The public deserves to know what's in the training data for #ChatGPT. But this is about transparency and accountability, not about measuring “intellectual contribution.”
@jhoang314 It's important to understand the dynamics of innovation. The reason why AI is progressing so fast is *precisely* because Nets & Google are quite open about publishing research and code. Imagine if Google & Meta filed tons of patents and started enforcing them...
It's hard to overstate just how important the right distribution channel can be to a product's success. GPT3.5 became available in May/June, but it wasn't until OpenAI released the chatbot half a year later that it took off overnight. https://twitter.com/...
OpenAI didn't comment for this column, or make anyone available for interviews. But I'm grateful to the people who did talk. I think we, as a society, need to know more about the people and companies that are releasing powerful AI tools into the world. https://www.nytimes.com/...
i think this helps explain why i'm so overwhelmed by generative AI tools...it's not just what they can do or the paradigms they could shift...it's the speed of adoption and the way they're adopted culturally (how there's something that feels lasting about it) https://twitter.com/…
“Two months after its debut, ChatGPT has more than 30 million users and gets roughly 5m visits a day. That makes it one of the fastest-growing software products in memory. (Instagram took nearly a year to get its first 10m users.)” https://www.nytimes.com/...